User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/79.0.3945.16 Safari/537.36 Edg/79.0.309.14

Steps to reproduce:

Open an app monitor that can monitor your gpu/memory gpu clocks like msi afterburner or HWinfo64, then go for example to youtube and jump continuously from one video to another.

Actual results:

You'll see the gpu clocks and gpu memory clocks jump at their maximum clocks (normal clocks not max boost clocks). In my case my rtx2070 go to 1410gpu clock, and 7000mhz memory clock. If you do this continuosly it will increase temp and power consumption a lot because it gpu won't ramp down its clocks. This happened aswell on my 1060 6gb. I noticed this issue happend even on twitch, but when on youtube this not happend on some old firefox build, on twitch it happend always, even on FF builds which works fine on youtube.

Expected results:

The gpu should increase its clocks only by just a little. This is how it works at the moment on chrome, edge chromium and some old FF version. I've run mozregression and tested it. Some version works fine. I can post only one file, but I have done two logs, one with "search for a bugfix instead of a regression" ticked and other without.

My full specs are Ryzen 5 1600 Asrock AB350M Pro 4, EVGA RTX2070 XC(driver 441.12), 16gb of RAM. Windows 10 Pro build 1903(18362.449).

Added even the other log

Range is: 2017-04-15, 2017-04-16 in the first log and 2017-06-09, 2017-06-10 in the second log.

jya, I see something from you in the second log. In any case, maybe it's because we do VP9 hw decoding using this RTX2070, and Chrome and others don't ? What codec twitch is using ?

Opening a random twitch stream I get "Codec: avc1.64002A,mp4a.40.2" I tried to open it both with firefox and chrome and the codec is the same.

Ok, clearly something that should use the GPU for decoding and chrome and edge probably do the same, thanks for the additional info!

Just for curiosity I tried to open the same youtube video both on chrome and firefox (this video: https://www.youtube.com/watch?v=wGDFm3hUY8s).
In both case the codecs showed was: vp09.00.51.08.01.01.01.01 (248) / opus (251).

This isn't something we control.

The 2070 has a hardware VP9 decoder, we use the VP9 MFT from Microsoft (same as Edge and Chrome). The difference however is that we are copying the GPU image immediately into another while recent version of Chrome which does the decoding on demand from the GPU compositing side doesn't have to. That copy is particularly inefficient with nvidia card (works okay with Intel)

I'm moving this to GFX for now. Where the problem stands is blurry. It would require a complete rewrite of our decoding/compositing stack.

This enhancement is tracked in bug 1539735. I'm not when we will be able to get to it though.

Just for give more information, these days I even tried with my old 1060 6GB with the same results: gpu clocks (base clock) and gpu memory clocks maxed out with same modality of my rtx2070 described above.